Related Experiment Video
Updated: Nov 5, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
736
Automatic colonic polyp detection using integration of modified deep residual convolutional neural network and
Win Sheng Liew1, Tong Boon Tang1, Cheng-Hung Lin2
1Department of Electrical and Electronic Engineering, Universiti Teknologi PETRONAS, 32610 Seri Iskandar, Perak, Malaysia.
Computer Methods and Programs in Biomedicine
|May 13, 2021
Summary
This study introduces an advanced artificial intelligence (AI) computer-aided diagnosis (CAD) tool for early colorectal cancer (CRC) detection. The novel AI model accurately identifies colonic polyps in endoscopic images, improving upon existing methods for enhanced diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Colorectal cancer (CRC) incidence and mortality rates are increasing.
- Current artificial intelligence (AI) computer-aided diagnosis (CAD) tools show promise but require improved sensitivity for polyp detection.
- Early detection of colonic polyps is crucial for reducing CRC mortality.
Purpose of the Study:
- To develop a novel CAD tool for accurate colonic polyp detection.
- To enhance the sensitivity and accuracy of AI-based polyp detection systems.
- To improve early diagnosis of colorectal cancer through advanced AI algorithms.
Main Methods:
- A modified deep residual network (ResNet-50) integrated with principal component analysis and AdaBoost ensemble learning was developed.
- Endoscopic images underwent preprocessing including median filtering, thresholding, contrast enhancement, and normalization.
- The model was trained on a combined dataset of 3 publicly available polyp image datasets (Kvasir, ETIS-LaribPolypDB, CVC-ClinicDB).
Main Results:
- The proposed AI approach achieved a Matthews Correlation Coefficient (MCC) of 0.9819.
- High performance metrics were recorded: 99.10% accuracy, 98.82% sensitivity, 99.37% precision, and 99.38% specificity.
- The model demonstrated robust performance across diverse datasets.
Conclusions:
- The developed AI method effectively classifies endoscopic images for polyp detection.
- This tool shows significant potential for developing advanced computer-aided diagnostic systems for early CRC detection.
- Automated and accurate polyp identification can lead to improved patient outcomes.
